Accurate and fully automatic 3D registration of spinal images using normalized mutual information

Automatic and accurate multi-modality (CT/MRI) image registration is an important part of image guided surgery, pre-surgery planning and post-surgery evaluation. Surface based registration is commonly used for registration of CT and MRI images of bone. Surface extraction from CT and MRI datasets is a pre-requisite for the registration. It is known that it is not possible to achieve fully automatic, accurate and complete segmentation of the spine from MRI dataset. Thus surface based registration for CT-MRI spine datasets cannot be fully automated. In this paper, we investigate the use of normalized mutual information as a method for fully automatic and accurate registration of CT-MRI spine datasets. We have compared the registration results with those from the surface based registration. Our results are promising and show that normalized mutual information can be used to implement fully automatic and accurate registration for CT and MRI images of the spine.